Machine learning as a tool to improve groundwater monitoring networks

Machine learning techniques are gaining recognition as tools to underpin water resources management. Applications range widely, from groundwater potential mapping to the calibration of groundwater models. This research applies machine learning techniques to map and predict nitrate contamination across a large multilayer aquifer in central Spain. The overall intent is to use the results to improve the groundwater monitoring network. Twenty supervised classifiers of different families were trained and tested on a dataset of fifteen explanatory variables and approximately two thousand points. Tree-based classifiers, such as random forests, with predictive values above 0.9, rendered the best results. The most important explanatory variables were slope, the unsaturated zone’s estimated thickness, and lithology. The outcomes lead to three major conclusions: (a) the method is accurate enough at the regional scale and is versatile enough to export to other settings; (b) local-scale information is lost in the absence of detailed knowledge of certain variables, such as recharge; (c) incorporating the time scale to the spatial scale remains a challenge for the future.